I've been arguing in the PokerStars bot thread

http://forumserver.twoplustwo.com/28...rstars-828016/

that it should be much easier to create a world class computer poker player (for NLHE 100BB cash games) than it was to create a computer program that beats the best chess players. The reason for my opinion is that computers need to use this evaluation function, which is not something that human players do in a mathematical way. Human players rely on intuition to evaluate many particular aspects of a particular position, so it is very hard to put that iinformation nto a compuiter for ANY position.

On the other hand, to beat onlike poker games, you just need to put into the program a (large) number of purely logical variables in exactly the same way human players think.

I think this is pretty clear. I give a few more details on the reasons for my opinion there, but some people seem to disagree. However, I think they are just the shortstack haters, so they just disagree for their own interest (bots can beat 20BB games but not 100BB games).

Do you agree with me or not? Why?

Bonus question: I've been curious for some time on the following question, that is intuitively clear. Is it possible to prove nowadays (i.e., without a computer that is able to find "the perfect chess game") that with perfect play from both white and black, white does not lose?

I don't think that even using computers this question can be answered. Anyone has an idea?

I saw that thread but there is no way on earth I am going to read all of it. Tell me the page numbers please

As far as your question.. I am not on your side In the absolute majority of positions brute calculation force > human intuition. It is simple as that, humans are making calculation mistakes while computers aren't.. Of course it still takes programming and other skills to create a program with such calculation force.

As a matter of fact, I think you got it reversed - the fact that there are so many unknown variables and various psychological aspects in poker makes it much much harder to create a profitable computer program for it. I am talking about 100bb+ of course, screw all those shallow tables.

For the bonus question - no, it is not possible to PROVE it, yet vast majority of people would answer this based on their intuition (me included). White shouldn't lose with perfect play from both sides

It all began in post #1222 in that thread (at around page 80 with the forum configuration).

Calculation is enough the beat terrible human players. In the late 80s most (or all) computer programs were just terrible because they couldn't consider anything farther than the number of moves they are able to calculate. so for instance, if they could calculate 5 moves and a pawn would promote in the 6th move they wouldn't consider it.

And I don't think there are really "pshycological aspects" to consider with respect to other players. Some people may believe they think that way, but I think it is just some statistics reflected in their HUDs plus recent history. All that information can be put into a computer.

To prove this from a game theory point of view would effectively require solving the games. So proof no.

But in soft terms all that would need to be demonstrated that moving first is not a disadvantage.

If you were to take the broken mirror strategy:

Copy your opponent's moves until such time that you think that there is a clear advantage to not do so.

This is a very irritating strategy so please don't play it against me


If you do this for chess the vulnerability of the king allows for white still to have clear tempo. This would suggest that white cannot be -EV. I do not know how closed the opening analysis is for chess, but the game may prove to be soluable but no one is yet close.

if it was easier, it would have been done by now.

Well, again I agree intuitively. But I don't think it is possible to PROVE that answering my question would require solving the game either.

There are many mathematical questions that have an answer that is not constructive. Proving that there is not such a non-constructive answer for my question seems difficult to me too.

Why? Do you know the number of years, the number of people working on it, and the amount of money devoted to it? Such has not been the case for poker.

Also, I'm quite sure that I've read than there are simpler games than chess (backgammon perhaps?) where computers are much worse than they are at chess, simply because people has not been working on those games

I'd already touched on my views in the original thread. For a perfect random number generator a computer program can easily be written to outperform a pro. Especially, if they had to play with the same username each time.

Why? - poker is basically a three move choice raise, fold and call /check. While the NL is not favourable to a computer. It is still just statistics to decide on which action to make. The computer can play unreadably but a Human will always give off some information. A computer will know precise odds for all situations whereas a player will use approximation.

And how do you make a bot use that information? Or, more importantly, how do you make it to ADAPT? All you can do is put various variables in it, something like "if 9/8 player 5bet shoves preflop fold". But how do you make him to adapt, I don't understand. The regs WILL over time notice those patterns of the bot and will adapt and use it, and what will bot do? Keep playing the same way cause he is a ****ing bot lol


Or are you trying to say that the bot will not need to adapt because he knows odds perfectly? facepalm.jpg

Backgammon computers proabably are weaker than computer chess playing programs but they have dominated backgammon for years.

I play several new games and it is very striking the difference between efforts produced.

One game for example where a strong program has been produced is abalone:

http://www.ist.tugraz.at/staff/aichh...Tournament.pdf

That became very strong when trained on the then top 3 players in the world.
With enough human games I think the heuristics become dominant but they reuse ideas from chess programming.

Other games where lesser efforts have been made fail to beat a beginner. With poker it is 2-fold no European University is researching nor any of the "IVEY" league in the US. With no fully published matches either it means that an amateur will not succeed in getting the strategy right.

It would take a lot less effort for poker heuristics than chess as poker is based upon reads and tells. The computer has no weakness in giving out tells so the top players lose their only edge over the computer.

I think that for many hands perfect odds gives a PC an edge.

For the rest the heuristics may need to be learned from how humans behave or basic in the dark probabilities: as he is raising 3 out of ten hands so his average hand is QJ, etc... With minor player adjustments.

But stratego offers a combination of chess and poker and in Holland is played to a very high level. There are strong programs in this.

The track record is that for a human to compete against the computer it requires being able to think through to the end and go beyond a computers 15 move search depth heuristics. Poker - just does not offer the scope to do this in cash games.

MTT is more challenging but there is not currently the evidence that pros are doing this perfectly, otherwise there would be many more top names in the last 200 for the WSOP. If I was going to write a bot this is what I would chose as it is more interesting.

No offense sir, you certainly are out spoken and have interesting ideas, but I have a feeling that you haven't played a lot of poker have you?

I agree that computers can avoid giving tells. They can randomize their play among all decent plays, just like human players can.

That is not the point at all. Ok, let's say you put an algorythm which says that in that particular situation with that type of hand the bot will shove 30% of time and call the other 70%. Human pro player will figure this out eventually, and **** bot based on this info until bot's owner makes some adjusments. BUT in order for him to even understand what adjustments are needed he has to be a great player himself, which means he could be easily making money himself without wasting that time on developing the bot.

I would be willing to bet big amounts of money that if you let Ivey or Durrrr or one of them other sick players play even the absolutely top of the line bot for a big enough sample without letting the owner of the bot to make adjustments, the bot will be CRUSHED.

Well, the randomize feature can include "preadjustments", i.e., it can pick a number of different top players and play each day like one of them. How would human players detect that? Obviously I'm not talking of anything a single person can do, just like any single person would have gone nowhere in chess.

I doubt anyone will be able to program a bot that can analyze the style of specific opponents and make optimizing adjustments. I think that such a bot would probably not be able to beat top poker players.

It is more likely that people will eventually program bots to play a style of poker that closely matches game-theory-optimal play and is therefore (practically) unexploitable. This would be much easier in certain games like Sit and Gos, Draw or Limit Holdem, but not impossible in NLHE cash games.

lol, do you know why no computer is yet able to beat GO grandchampion on an 8x8 fast table?

just visualize how you would play a hand like AK in a tree. draw all different scenarios, for each flop, turn and river. consider stacksizes, the amount of players in the hand, reads, tendencies, all betting patterns (donkbet, overbetshove, checkraise etc) after you completed that wall of paper, think about what the bot should do with other hands. than develop an algorithm that can detect a player capable of adjustments and a hand played by your bot that would lead detected player to adjusting. write a tree of adjustments for all scenarios. make the tree applicable to multiway pots.

then beat a worldclass player.

I think 8x8 Go is a much simpler game than chess (in terms of the total number of possible games) and strategical ideas involved . So the only reason i can think of is that much less effort into making such a computer program has been done than it was done with chess.

complete chess noob: Shouldn't WHITE have the advantage since it acts first and therefore can move the pieces out first?

Oh and, lol you guys don't play a lot of poker, do you? A bot can't win just because he always knows the odds, or never gives off tells.

No, the key to beating poker (and especially HU) is ADAPTION. You need to constantly adapt to the opponent player. How he's thinking, how he adapts to what you just did, and counter-adapt that. So if a computer would ever beat a human player, it would have to be able to adapt, so I guess AI is the only way to go? So when advanced AI computers come out, THEN i'll be afraid

Intuitively that's obvious to everyone. However, there are positions were the side that has to move first loses (which is called reciprocal zugzwang). Everyone is sure that the initial position is not one of these ones, but proving it mathematically is a different issue.

This adaption, in online poker, basically means paying attention to the HUD data and the recent history, so that you can evaluate what he is doing. I don't see why you cannot put that into the computer.

In addition, there is no way human players can adapt to the computer if the computer has many completely different styles. Imagine that it tries to imitate the style of 100 top players and there is a random procedure to choose how to play any given hand.